The Voiceprint Identification Systems Based On Optimization Of Voice Extracted

Voiceprint identification has a significant impact for people’s lives, so effectivevoiceprint identification system is necessary. As for the recognition rate determines theoverall system, the most critical in the endpoint detection are feature extraction andtemplate matching, in order to achieve better recognition results, this paper studies theendpoint detection and feature extraction impact on the overall performance of thesystem, at present, a joint of variety of characteristic parameters has become animportant optimization parameters way.This text firstly started in the preprocessing stage of voice signal, removed thevoice signal noise, detected the endpoint, removed the silent segment of voice signal,and provided the speech features collection an effective speech segments.Comparingthe detection performance of Cepstrum algorithms and Logarithmic energy featurealgorithm, the results showed that,the two detection performance are not very well inlow SNR conditions, so in order to analyze voice signal better, we must find new waysto do the endpoint detection.In order to overcome the weak points of Cepstrum distance voice detectionalgorithm, this text combined the logarithmic energy (LE) Character and the Cepstrum(C) Character, and proposed a new kind of method, that was logarithm energy Cepstral(LEC) Character.the method used fuzzy C-means clustering and Bayesian InformationCriterion (BIC) to estimate and determine the characteristic threshold, and achieved acorrect and effective voice endpoints judgment.Among three very typical noise,simulating the speech with noise at the station of Signal-to-noise ratio from-5dB to15dB,experimental results showed that the detection error rate of the LEC method isjust20.25%, significantly lower than the cepstrum and logarithmic energy method. This method indirectly improved the voice recognition results and was more effective fordetermining the endpoint of the voice.In this text, we used MATLAB dedicated voice processing toolbox, extracted thecharacteristic parameters of the input voice clips, matched the reference template andtest templates by dynamic time warping algorithm, and improved the systemidentification rate. In this text, we considered to the reliability of the system, let theMFCC number as the characteristic parameters of the voice clips, and provided atheoretical basis for the system’s wide range applications. It has the advantages such asa fast speed of operation, easy to update the template, and a small amount of calculationas well as low error rate and so on.In order to compare the various types of endpoint detection algorithms better, thistext also achieved the log energy features and Cepstral detection algorithm. This text letthe Mel Cepstrum difference as the characteristic, achieved an ideal recognition rate,and so that it can be used in the occasions where the demanding recognition rate is notvery high.